Cortex is a CLI based memory layer which correctly stores a user’s notes (i.e markdown or txt files), it lets users store relevant information from the note which is also consulted to a LLM and finally into the database and now users can easily with the help of cortex MCP ( & skill to instruct) allow agents to bring in context regrading a process or idea or just a personal preference, anything into the current session with an agent and use that to work on the task.
Challenges Faced :
The core package was the hardest to build because it laid the foundation for everything: the graph, the entities, keyword search, nearest facts, ingestion, LLM suggestion, embedding, status & timeline all these features are the main features on which the entire app, the CLI and the MCP is built on. Every other package is just interacting with these core utilities.
Redis was new to me. I knew that it helps in rate limiting, but while building this app, I found out it's not a rate limiter, it is a data store in of itself also helps with background jobs or queuing which workers can pick up. Redis made sense when you accidentally exit the terminal while a process like trying to ingest a page gets interupted and that's the moment when we need Redis working its magic so that when you come back you can resume from where you left that same job, will get resumed.
Tech Stack :